A Framework for Business Deployment of Machine Learning Models under Delayed Prediction Results

Abstract

Recently, many predictive models of machine learning algorithms have been deployed in real business works. In such a situation, issues from the deployment of machine learning models are discussed at length under 'ML Ops'. We discuss that an important practice of ML Ops cannot be applied to models used in the retail stores where the results of prediction are often achieved after a certain period. In this paper, we propose a framework for such a condition of delayed prediction results, to stimulate further discussion. With this framework, we conclude that one should monitor the data drift and business operators should play a role beyond a user.